fix(weixin): add prefer_platform_transcription config option

[answerallen]

What does this PR do?

Adds a prefer_platform_transcription config option to the WeChat (Weixin) adapter that lets users choose between Tencent Cloud's STT and Hermes' own local STT pipeline for voice messages.

When true: trust Tencent's voice_item.text as the message body, skip downloading raw audio.
When false (default): always download raw audio and route through Hermes' local STT pipeline (unchanged from #27300).

Problem

After #27300 was merged (PR #73515), Hermes always ignores Tencent Cloud's voice_item.text and downloads raw .silk audio for local STT re-transcription. This fixed garbled transcriptions for non-Chinese users but broke Chinese-heavy users who:

• Have stt.enabled: false (no local STT provider configured — no faster-whisper, no GROQ_API_KEY, etc.)
• Were previously relying on Tencent Cloud's high-quality Chinese STT
• After upgrading, voice messages become a placeholder ([The user sent a voice message: ...]) with no transcript at all

This is a common setup for Chinese users on low-resource servers where installing faster-whisper is impractical (300MB+ model, OOM risks on ≤2GB RAM).

Solution

Add prefer_platform_transcription to gateway.platforms.weixin.extra:

gateway:
  platforms:
    weixin:
      extra:
        prefer_platform_transcription: true

Changes

gateway/platforms/weixin.py:

• _extract_text(): new prefer_platform_transcription keyword arg. When true, returns Tencent's voice_item.text as the message body (with [Voice transcription provided by Weixin] prefix from fix(weixin): preserve voice transcript origin #65022).
• _download_voice(): when prefer_platform_transcription is true and Tencent text is available, returns None (skip audio download).
• WeixinAdapter.__init__(): reads extra.prefer_platform_transcription (default: false).

tests/gateway/test_weixin.py:

• New TestWeixinPreferPlatformTranscription class with 8 tests covering:
  • Text extraction with/without the flag
  • Audio download skip/download behavior
  • Config coercion

Behavior matrix

prefer_platform_transcription	Has Tencent text	Has raw audio	Result
false (default)	yes	yes	Download audio, ignore Tencent text (#27300 behavior)
false (default)	yes	no	Use Tencent text (nothing else available)
false (default)	no	yes	Download audio, local STT
true	yes	yes	Use Tencent text, skip audio download
true	no	yes	Download audio, local STT (fallback)
true	no	no	Empty (nothing available)

Related Issue

Related to #27300, #65022. Does not revert #27300 — adds an opt-in escape hatch for users who want platform-side transcription.

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests (adding or improving test coverage)
• ♻️ Refactor (no behavior change)
• 🎯 New skill (bundled or hub)

How to Test

1. Set gateway.platforms.weixin.extra.prefer_platform_transcription: true
2. Send a Chinese voice message via WeChat
3. Confirm the agent receives [Voice transcription provided by Weixin]\n<transcript>
4. Set it to false (or remove), send the same voice message
5. Confirm audio is downloaded and local STT is used instead

Automated:

scripts/run_tests.sh tests/gateway/test_weixin.py -v

Checklist

• I've read the Contributing Guide
• My commit messages follow Conventional Commits
• I searched existing PRs and issues
• My PR contains only related changes
• Default behavior is unchanged (opt-in only)